Color Infrared to True Color

In the past few years, digital photo data have become much more available via the Internet (Terraserver being the most well-known example). Nearly all of these images are panchromatic (black and white), since color is more difficult to produce and generally not needed for GIS tasks. Sometimes, however, it's very nice to have a good color aerial photo. This usually requires a custom flight and is very expensive.

Is there any alternative if you're looking for color imagery? Perhaps, depending on how you define "color." Many states now have Color Infrared (CIR) photo data available. An example of this type of image is shown below. Data provided by the Platte River Program.

Color Infrared images are created not so much for their color accuracy or aesthetic value, but because they are so useful for scientific analysis. Healthy vegetation is a very good reflector if Near Infrared (NIR) energy, and this is shown as bright red in the CIR image.

Aside from some sort of complex or bizarre image process, is there any way to simulate natural color using this image? To answer that, let's first take a closer look at Color Infrared film and how it works. Examine the diagram below.

This diagram shows how energy components from the scene are translated to the final CIR image. Each pixel in a computer image is composed of a red value, green value, and blue value. Since our eyes cannot see NIR energy, the way to represent it on an image is to translate the NIR value to something we can see-- in this case, red. So parts of the CIR image that appear very red are in fact strong reflectors of infrared energy.

You can also see from this diagram that red from the scene translates to green on the image. Thus, parts of the image that have a greenish tint are in fact red in real life. In a similar way, objects that are green in real life will apear to be blue on the CIR image. But you'll notice that there's nothing in the image that captures blue; any information about blue in this image is lost forever. Objects that are pure blue in real life will in fact appear black on a CIR image.

To create a true color image, we need the actual red, green, and blue components. What we actually have to work with are the NIR, red, and green parts. Let's go to the imaging software and see if it's possible to improve this image and make it closer to natural color.

Above you see a full-resolution view of part of the Color Infrared image. Let's first split this into its component red, green, and blue values.

From left to right, these are the brightness values for red, green, and blue (which represent Infrared, Red, and Green, respectively). To avoid confusion, we can refer to them as Channels 1, 2, and 3. Using simple imaging processes, it's possible to recombine these in any way to create a new color image. Since plants appear very bright in the infrared channel (1), it might work to map this to green to bring out the vegetation. Channel 2 (actual red) can be used for the red component. We have no data for blue, but it can be faked using the actual green channel (3). The result of the combination is shown below.

This is certainly a better image, in a lot of ways. It replaces the unnatural red tint with a nice green. One of the major flaws is that the green is too bright and looks unnatural. Another is that non-vegetation features (roads and houses) don't have much color and are tinted purple.

For the next try, we'll just forget about the Infrared information. Channel 2 for red and Channel 3 for green we know to be accurate. So for the next combination, we'll use those two and again fake blue by using Channel 3.

This is not appealing at first glance, but take a closer look. The roads are about the right color, and a few of the houses show promise. Also, the tennis courts look almost perfect. But most of this image is nearly black. For various reasons, reflectance in the true green channel (3) did not work out as it should have. The vegetation needs to be brought out more to correct this. Fortunately, we still have the NIR channel to work with.

Below you see the results of a simple average between Channels 1 and 3. Most of the true green information (Channel 3) is preserved, but the Infrared (Channel 1) is mixed in to bring out the grass and trees. The result is a mixture of the two, which should represent green well in the final image.

Now onto the final image, using the above mixture as the green channel.

It has its flaws, but this is about as close as we're going to get without taking an actual photo. When this is expanded to the larger scene, here is the image we get:

The above image had a bit of retouching (mostly to correct for water, since it absorbs nearly all Infrared radiation and appears black on the NIR channel), but it was a mostly automatic process using the above steps. For missing the entire blue section of the spectrum, this is a pretty good representation of the area. For an idea of a few flaws to this technique, see the image below.

Here you see two significant flaws. First is that the image seems overly biased towards green, and the field on the left looks washed out. Second, note that the shopping area on the right isn't very colorful. Its monochrome look is probably due to our lacking the blue color data.

That's all for this tutorial. Translation of CIR to True Color may not always be the answer, but this process should work well for most urban areas. Some terrain images of Kearney (created by overlaying this texture onto elevation data) are now in the Kearney Image Gallery. Good luck with your own creations!